In a k8s env (or other cloud), it is possible that some host entries in system.peers are invalid. e.g. rpc_address and/or host_id nulled.
In those cases, the endpoint creation fails and raise an exception. The good behavior would be to simply ignore those invalid hosts. The metadata topology monitor should be responsible for checking the host row validity.
The same logic than the java driver should be applied for this need. If one of the following fields are null, log a warning and ignore the discovered node: rpc_address, host_id, data_center, rack and tokens.
What about other fields from system.peer? (See JAVA-852)
This is indeed what was implemented. https://github.com/datastax/python-driver/pull/1092
Cool thanks. I missed that PR.